Predicting the global far-infrared emission of galaxies

Dust absorbs stellar emission and reradiates this energy in the far-infrared (FIR). FIR observations hence give us a direct view of the dust, and allow us to study its properties. Unfortunately, FIR observations are only available for a small subset of galaxies. In this work, we estimate the global FIR emission from global UV-NIR observations. We show that a machine learning method clearly outperforms a SED modelling approach. For each galaxy, we not only predict the FIR flux across the 6 Herschel bands, but also estimate individual uncertainties. We inspect the worst predictions, and investigate how the machine learning predictor generalizes on new data. Our predictor can be used as a virtual observatory, which is especially useful now that there is still no confirmed next-generation FIR telescope.